Bubbles | On Unpredictability and the Work of being Human | Something Entertaining

In This Issue:

• Bubbles. Are We? Aren't We? Read more below for my current take...
• On Unpredictability and the Work of Being Human. Helen explores one of the most important—and overlooked—dimensions of our relationship with AI: that unpredictability isn’t a bug of being human, it’s the feature.
• Something Entertaining: Cabin in the Sky. More below...

Bubbles

The world feels awash in talk of bubbles. Everyone is suddenly a market historian, a macro strategist, or an amateur Fed whisperer. At the Artificiality Institute, we’re not in the prediction business—and honestly, we don’t want to be. But I have spent 25 years watching technology cycles up close: first on Mary Meeker’s internet research team during the dot-com build-up, then a decade later as a cleantech analyst watching that sector rise, fall, and reinvent itself. That experience gives me a certain perspective on today’s AI moment.

Before asking whether AI is a bubble, it’s worth looking at the water we’re swimming in. Markets today are pricing relative certainty in a world of absolute uncertainty. Every day, hundreds of millions of shares in the “Magnificent Seven” change hands. Those trades produce what we call “the market’s view,” but a stock price isn’t only a view on a company—it’s a view on everything else in the economy. And right now, everything else looks shaky.

Strip out AI- and data-center-related investment, and the U.S. economy would be flirting with stagnation. Analyses suggest that most of the growth in the first half of 2025 came from data centers and related high-tech spending; remove those and GDP growth falls close to zero. Consumer sentiment, as measured by the University of Michigan, has fallen back toward record lows. The headline index hovers near 50, far below its long-run average, with the “current conditions” sub-index hitting its lowest level in the survey’s 73-year history. People feel it. They’re anxious about work, frustrated by prices, and deeply skeptical that anyone in charge has a real plan.

At the same time, unemployment has drifted into the mid-4% range—its highest in several years—while inflation remains above pre-pandemic norms. That leaves the Federal Reserve juggling an economy that looks softer and price rises that haven’t fully relented. Add a federal government that has turned into an uncertainty machine—shutdowns, fiscal standoffs, regulatory whiplash—and you lose one of the traditional anchors of U.S. economic strength: policy predictability.

Put bluntly, there isn’t much for investors to put money into except tech. Tech offers predictable margins, durable demand, and business models that don’t collapse when unemployment ticks up a bit. Large tech firms are also skilled at managing their relationships with policymakers, which lowers the risk of surprise regulation. In a fearful world, markets crowd under the one awning that looks sturdy. Right now, that awning is tech—and AI is the biggest growth story under it.

So is AI a bubble? Let’s look at the main arguments.

The first claim—“AI stocks are wildly overvalued”—isn’t well supported by the numbers. Alphabet, Meta, Amazon, Microsoft, and even Nvidia trade at earnings multiples that are elevated but not historically abnormal. Apple is the outlier, sitting at a decade-high multiple, but Apple isn’t really an AI company today. (As for Tesla… I’ve stopped trying to understand Tesla.)

A second claim is that private AI valuations are “insane.” Maybe. But this is venture capital: many failures, a few big successes, and heavy dispersion in between. Froth at the edges tells us almost nothing about the long-term value of the category. High dispersion is a feature of this style of investing, not a bug.

A third claim—the most serious one—is that tech is overbuilding data centers. Here, the telecom comparison is useful, but not in the way it’s usually told. In the late 1990s and early 2000s, telecom companies invested roughly $2 trillion (in today’s dollars) laying fiber and building network infrastructure. The problem wasn’t just the scale—it was utilization. Much of that fiber sat dark for years, generating no return and dragging companies into bankruptcy. They built ahead of demand.

AI infrastructure looks similar in scale, but very different in how it’s being used. Today’s data centers are already maxed out: GPUs are fully booked, lead times are long, and prices are rising. The assumption behind the current build-out is that every new facility will be consumed by real, existing demand. That’s the opposite of the telecom boom.

None of this means AI investment is risk-free. I have one primary concern: the market may be mispricing where AI value will reside. Most current valuations assume a future in which intelligence is centralized, metered, and monetized through the cloud—every query an API call to a data center. Centralized capacity dominates today because it’s the only place large models can realistically run. But that constraint is temporary. As models become smaller, faster, and more capable, the center of gravity will start to move away from the cloud.

But we expect a significant share of inference to migrate to the edge: small, powerful models running on your phone or laptop; real-time context pulled directly from your device; inference costs approaching zero; and far better privacy, autonomy, and cognitive sovereignty.

If that plays out, the cloud-first economics underpinning today’s valuations will weaken. Not collapse, but shift. Value will migrate toward companies that can design, deploy, and integrate AI into everyday devices and cognitive routines—not just those that own the largest clusters.

So are we in a bubble? Based on the simple fact that current demand is outstripping supply (just note how often your use can be throttled), probably not. But it's certainly possible that some may be pricing in the wrong assumptions. The cloud-first, compute-metered story of AI is the version the market currently believes. If intelligence shifts from the cloud to the edge—from centralized systems to personal, contextual models—the locus of value will move as well and the centralized compute will be used for sources of demand that are different from today.

This all might mean the value may accrue in different places—and in different companies—than today’s markets expect. And that, more than any chart of price/earnings ratios or capex forecasts, is the piece of the bubble conversation that’s worth paying attention to.

On Unpredictability and the Work of Being Human

Helen explores one of the most important—and overlooked—dimensions of our relationship with AI: that unpredictability isn’t a bug of being human, it’s the feature. As AI systems become better at modeling our habits, preferences, and even our reasoning patterns, the instinct is to assume that the value of being human gets squeezed into a smaller and smaller corner. But Helen argues the opposite.

Unpredictability is where our agency lives. It’s how we create new meaning, imagine new futures, and exercise judgment in situations that don’t fit neatly into a pattern. AI is powerful precisely because it predicts; humans are powerful because we transcend prediction. The space between these two dynamics—the stable logic of machines and the improvisational logic of people—is where the real work of the future happens.

This piece is a reminder that the goal of AI isn’t to eliminate uncertainty, but to give us more room to move within it. The more capable our tools become, the more essential our unpredictability is to the systems we build and the lives we lead.

Read more...

Something Entertaining: Cabin In The Sky

As a fan of 90s era Native Tongues music, I was stoked to a new De La Soul album—especially after the passing of one of the original three, Dave "Trugoy" Jolicoeur in 2023. This album honors Dave with the clear statement that 3 will always be the magic number. De La's music may have mellowed over the years—and their humor might be appropriately muted—but it's still De La and that's awesome.

In today's world, it's interesting to hear them refer back to their 2000 album, Art Official Intelligence by saying they will always make art that's official and full of intelligence—or A.O.I. And among the many guest artists, De La includes a voice called H.O.P.E, or their Host Operating Program Entity. Their semi-dismissive attitude to H.O.P.E makes me wish we could get Maseo and Posdnuos on the podcast to talk about what they really think of AI and music.